In affluent countries where combination anti-retroviral therapy (CART) is widely used, life expectancy with HIV has increased well beyond 10 years, but this has greatly increased the number of people living with HIV/AIDS. 40 million people are now at risk for progressive HIV-related damage to the brain (Navia 2011), and 40% show some neurological or cognitive impairment. Building on our recent discoveries and innovations in MRI and diffusion tensor imaging, our project charts the dynamics of HIV disease in the brain, revealing factors that predict clinical decline and brain decline. By analyzing data from the HIVNC (HIV Neuroimaging Consortium) cohort (218 HIV+ subjects scanned longitudinally with MRI every 26-32 weeks for 3 years; 900 scans), we will use our sensitive image analysis method, tensor-based morphometry to (1) map rates of brain tissue loss over time, determining which brain systems lose tissue fastest; (2) relate these loss patterns to neurocognitive impairment (NCI); and (3) determine which host and disease-related factors, at baseline, predict higher atrophic rates over the 3-year follow-up interval. To deeply probe the causes of dysfunction, we will use whole-brain tractography based on diffusion tensor imaging in 267 subjects to determine where HIV+ patients have reduced fiber integrity. We will use each patient's DTI scan to compute a whole-brain connection matrix, based on a state-of-the-art whole-brain tractography method we developed. We will combine the best neuroimaging measures from Aims 1 and 2 into our support vector machine method to predict (1) future rates of atrophy, and (2) cognitive decline over the 3-year follow-up. With our best predictors of decline from Aims 1 and 2, we will predict which HIV+ patients will show imminent decline. We will estimate the sample sizes needed for a neuroimaging-based drug trial to detect a 5%, 10%, or 25% slowing in the rate of atrophy, and the same percents of slowing in the rate of cognitive decline. We will test whether our predictors generalize to the large, independent Charter and Miriam datasets beyond HIVNC (see Support Letters from Drs. Igor Grant and Ron Cohen), and when used by our many HIV research collaborators now using our methods; see Pilot Data). As shown in new pilot data, we assess how imaging protocol differences affect the measures; our innovations to reduce scan protocol confounds (e.g., adjusted FA) will guide selection of the most robust predictors. We will evaluate how useful these new measures are for predicting cognitive decline, and boosting power in an antiretroviral drug trial. In other words, to what extent can a DTI scan, and an MRI-based map of atrophy, help to make clinical predictions of imminent cognitive impairment? Can they help select a sample for a drug trial? Our activities will make clinical trial design more efficient by selecting subjects with greater potential to respond t future therapies. As always, we will disseminate our methods to both experts and trainees in medicine, neuroscience, engineering, and to our network of over 100 collaborating labs (including two national HIV consortia: HIVNC, CHARTER, and the Miriam HIV cohort), to accelerate their work.